Google has made a significant push into the agentic AI era with a series of announcements at Cloud Next '26, held April 19–25, 2026. The updates span new hardware, distributed training techniques, and expanded AI capabilities for enterprises.
TPU v8i and v8t Google introduced its new TPU v8 line, with the v8t training chip delivering 121 ExaFlops of performance and the v8i inference chip offering an 80% improvement in performance-per-dollar compared to previous generations.
Decoupled DiLoCo DeepMind's Decoupled DiLoCo enables distributed training of large language models across multiple regions with low bandwidth requirements, making it feasible to train models across geographically dispersed data centers without high-speed interconnects.
Gemini Enterprise Agent Platform Vertex AI has been rebranded as the Gemini Enterprise Agent Platform, supporting over 200 AI models. New Gemini API updates include Deep Research and Deep Research Max, which feature advanced reasoning and native chart generation.
Multimodal Embeddings and More Gemini Embedding 2, now generally available, supports multimodal enterprise AI across text, image, audio, and video.
Google's announcements signal a major shift toward autonomous AI agents and scalable infrastructure designed for enterprise-ready intelligence.